After Uber burned through its annual AI budget in roughly four months and imposed a $1,500 per month cap on per employee AI spend, Amazon and Meta are quietly walking back the 'use AI everywhere' mandate — dropping the AI token leaderboards and performance review weighting that…
Uber burned through its annual AI budget in roughly four months. The response, a $1,500-per-month cap on what each employee could spend on AI tools, is quietly becoming the template for a larger reckoning inside enterprises that spent the last year mandating AI use without seriously accounting for what it cost. That cap was reported by TechCrunch on 2026-06-02 and surfaced again in Futurism's recent aggregation of the cost-reckoning beat.
That reckoning is the underreported story behind the current coverage. The framing in the Futurism piece, which leans on The Economist's 2026-06-14 reporting, treats it as a CEO walk-back. A more accurate read is that the industry is moving from a "use AI everywhere" mandate phase into something closer to a budget discipline, with unit economics, usage caps, and a sharper question: where does AI actually pay for itself?
The receipts are concrete. The Ramp AI Index, covered by TechCrunch on 2026-06-10, puts per-employee AI spend at roughly $7,500 per month for the most AI-intensive businesses. That figure is the median of a small group of heavy users, not an industry average, but it is the benchmark executives are now staring at. At the visible edge, one employee was reported to have spent $150,000 in a single month on AI usage, per a March 2026 New York Times feature cited in the same aggregation. An Nvidia executive separately acknowledged that he spent more on AI compute for his research team than on the researchers themselves. None of these numbers are typical. All of them are signs of a distribution that is expensive well past the median.
Amazon and Meta have both dropped their internal AI-token leaderboards — ranking employees by AI usage volume — per the same Futurism-curated reporting. Meta had also factored AI usage into performance reviews. Both changes are small operationally, but they tell the cultural story: leadership no longer wants employees gaming a usage metric, because the metric was driving the wrong behavior.
OpenAI has reportedly been considering rate cuts to head off a price war with Anthropic, per Futurism's prior reporting. Token prices today are effectively subsidized to acquire customers, and the long-run question — raised in a New Yorker essay on AI profitability — is whether the unit economics ever close.
The next competitive edge in the enterprise will go to the organizations that can answer, with evidence, where AI actually pays for itself. The companies that lose this round will be the ones still mandating usage without measuring return.